Parametric methods for confidence interval estimation of overlap coefficients

详细信息    查看全文

• 作者：Dan Wang^a ; ^b ; Lili Tian^a ; ^{ltian@buffalo.edu" class="auth_mail" title="E-mail the corresponding author}
• 关键词：High-throughput platforms ; Genomic biomarker ; Overlap coefficient ; Generalized inference ; Parametric bootstrapping ; Mixture Gaussian ; EMEM algorithm
• 刊名：Computational Statistics & Data Analysis
• 出版年：2017
• 出版时间：February 2017
• 年：2017
• 卷：106
• 期：Complete
• 页码：12-26
• 全文大小：653 K

文摘

Overlap coefficient (94731630202X&_mathId=si125.gif&_user=111111111&_pii=S016794731630202X&_rdoc=1&_issn=01679473&md5=0cfe36530503bfe239ba1fbed0b1649e" title="Click to view the MathML source">OVL$OVL$), the proportion of overlap area between two probability distributions, is a direct measure of similarity between two distributions. It is useful in microarray analysis for the purpose of identifying differentially expressed biomarkers, especially when data follow multimodal distribution which cannot be transformed to normal. However, the inference methods about 94731630202X&_mathId=si125.gif&_user=111111111&_pii=S016794731630202X&_rdoc=1&_issn=01679473&md5=0cfe36530503bfe239ba1fbed0b1649e" title="Click to view the MathML source">OVL$OVL$ are quite sparse. This article proposes two methods, a generalized inference (94731630202X&_mathId=si127.gif&_user=111111111&_pii=S016794731630202X&_rdoc=1&_issn=01679473&md5=f7959f9ceae1379589d4f5fca61a9d75" title="Click to view the MathML source">GI$GI$) approach and a parametric bootstrapping (94731630202X&_mathId=si128.gif&_user=111111111&_pii=S016794731630202X&_rdoc=1&_issn=01679473&md5=ec2c82e6b6e2cba17dc4b37f21b1371c" title="Click to view the MathML source">PB$PB$) method, to construct confidence intervals of 94731630202X&_mathId=si125.gif&_user=111111111&_pii=S016794731630202X&_rdoc=1&_issn=01679473&md5=0cfe36530503bfe239ba1fbed0b1649e" title="Click to view the MathML source">OVL$OVL$ under the assumption of normality. In conjunction with the 94731630202X&_mathId=si130.gif&_user=111111111&_pii=S016794731630202X&_rdoc=1&_issn=01679473&md5=a0a137a2371c79584e7e651832f34831" title="Click to view the MathML source">EM$EM$ algorithms, these methods are extended to mixture Gaussian (94731630202X&_mathId=si131.gif&_user=111111111&_pii=S016794731630202X&_rdoc=1&_issn=01679473&md5=0bba45d2dafca6024457245d1c6d17e9" title="Click to view the MathML source">MG$MG$) distributions. The performances of these methods are evaluated empirically under a variety of distributions including normal, gamma and mixture Gaussian. At last, the proposed approaches are applied to a published microarray dataset from a gene expression study of three most prevalent adult lymphoid malignancies.